Comparative Study of Frailty Assessment Measures in Predicting All-Cause Mortality: Insights From NHANES

The 3 most frequently utilized frailty assessment measures are the Fried criteria, FRAIL scale, and Frailty Index (FI). This study aimed to compare predictive capabilities of these 3 measures regarding all-cause mortality in the United States and to identify the key predictive variables.

Cross-sectional study.

From the National Health and Nutrition Examination Survey (NHANES) 2005-2018 cycles, a total of 39,631 participants aged 20 and older were included.

Fried status, FRAIL status, and FI status were determined for each individual based on the cutoff values from the continuous scores of their respective scales. Univariate and multivariate models, incorporating 11 covariates-sex, age, body mass index, ethnicity, education, marital status, smoking status, alcohol intake, employment status, poverty-to-income ratio, and total energy intake-were fitted using Cox proportional hazards and 2 machine learning models. Model performance was assessed through Integrated Brier Score (IBS), concordance index (C-index), and area under the curve (AUC) values from 10-fold cross-validation. Key variable analysis was performed using permutation importance and C-index increment. Subgroup analysis was developed according to age.

In univariate analyses, FI consistently outperformed Fried and FRAIL, showing significantly lower IBS, and higher C-index and AUC values. In multivariate analyses, few significant differences were found. Permutation importance analysis identified age as the most important variable, followed by Fried status and FI status. Similarly, in C-index increment analysis, age was the top one variable. Subgroup analyses showed that FI status consistently performed best in all metrics across univariate analyses at least in 40-59 and 60-79 age groups. FI status consistently emerged as the most important variable in permutation analysis across all age groups.

FI demonstrated the best performance as a single predictor in predicting all-cause mortality, with age being crucial for enhancing predictive performance. Future research should explore the applicability of FI in different populations and its relationship with cause-specific mortality.